One-Dimensional Ising Model and Transfer Matrix Method

The Ising model on a one-dimensional lattice is the simplest interacting many-body system that can be solved completely, and its solution carries a lesson far larger than the model itself. Each site holds a spin $s_i=\pm1$ ; the energy is built from nearest-neighbour bonds $-Js_is_{i+1}$ and a coupling $-hs_i$ to an external field. Because the energy is strictly local — one bond at a time — the Boltzmann weight factorizes into a product of identical $2\times2$ matrices, one per bond. The partition function becomes a matrix trace, and the entire thermodynamics is governed by the spectrum of that single small matrix: the transfer matrix. The exact answer delivers a sharp, historically decisive conclusion. A one-dimensional system with short-range interactions does not order at any nonzero temperature. The magnetization is zero in zero field for all $T>0$ , correlations decay exponentially with a finite correlation length, and every thermodynamic function is a smooth, analytic function of temperature and field. This is the cleanest illustration of a general theorem (Landau, van Hove): one-dimensional systems with finite-range interactions have no thermodynamic phase transition, because the free-energy cost of a single domain wall is finite while its entropy grows with system size. Ising found this in 1925 for his thesis and — reading it as a failure of the model — abandoned physics; the deeper meaning, that dimension is decisive, took decades to appreciate. The transfer matrix is far more than a calculational trick. It is the discrete-time analogue of the quantum evolution operator $e^{-\tau\hat H}$ : the classical 1D chain maps onto the imaginary-time history of a single quantum spin, a $0+1$ -dimensional quantum problem. The largest eigenvalue gives the free energy; the ratio of the two eigenvalues sets the correlation length, $\xi^{-1}=\ln(\lambda_+/\lambda_-)$ ; and the degeneracy of the top eigenvalue is the transfer-matrix fingerprint of broken symmetry. In one dimension the top eigenvalue is nondegenerate for every $T>0$ , so order is pushed to $T=0$ — while in two dimensions the same machinery, now with an infinite matrix, produces Onsager's finite-temperature singularity. We solve the periodic chain cleanly, compute correlations and susceptibility, extract the low-temperature scaling of $\xi$ , and connect the result to the Peierls argument and the lower critical dimension.

Key takeaways

The 1D Ising model with nearest-neighbour coupling $J$ and field $h$ is solved exactly by writing $Z$ as a trace of powers of a $2\times2$ transfer matrix $T$ .
For a periodic ring of $N$ spins, $Z_N=\mathrm{Tr}\,T^N=\lambda_+^N+\lambda_-^N$ ; in the thermodynamic limit only the largest eigenvalue $\lambda_+$ survives.
Because $\lambda_+(\beta,h)$ is analytic for all $T>0$ (the off-diagonal elements are strictly positive), the free energy has no singularity: NO finite-temperature phase transition in 1D.
At zero field the correlation decays exponentially, $\langle s_0s_r\rangle=(\tanh\beta J)^r=e^{-r/\xi}$ , with $\xi=-1/\ln\tanh\beta J$ diverging only as $T\to0$ .
The Peierls balance explains why: a single 1D domain wall costs finite energy $2J$ but gains entropy $\sim k_B\ln N$ , so disorder always wins at $T>0$ — the lower critical dimension of the Ising model is one.
Key idea: A bond-local Boltzmann weight makes $Z$ a trace of a product of identical $2\times2$ matrices; the largest eigenvalue $\lambda_+$ is analytic for all $T>0$ , so the 1D Ising chain orders only at $T=0$ , and the correlation length is set by the ratio of the two eigenvalues.

The core insight

A bond-local Boltzmann weight makes $Z$ a trace of a product of identical $2\times2$ matrices; the largest eigenvalue $\lambda_+$ is analytic for all $T>0$ , so the 1D Ising chain orders only at $T=0$ , and the correlation length is governed by the ratio of the two eigenvalues.

Where this lives today

The 1D transfer matrix is a workhorse well beyond textbook Ising chains. In biophysics the two states of a residue — helix or coil, base-paired or open — map onto Ising spins with a nearest-neighbour cooperativity $J$ and a field $h$ set by the local free-energy difference; the transfer matrix produces the sharp-but-not-singular helix–coil and DNA-melting curves $\theta(T)=\tfrac12(1+m)$ , whose apparent first-order look is just finite-cooperativity rounding of a system that, strictly, cannot have a true 1D transition. The exact solution also establishes the one-dimensional Ising model as the canonical example of a system at its lower critical dimension: the Ising chain fixes $d_\ell=1$ for discrete symmetries, the discrete cousin of the Mermin–Wagner theorem ( $d_\ell=2$ for continuous symmetries). It is the standard benchmark against which mean-field and Bethe–Peierls approximations are calibrated — both wrongly predict a finite $T_c$ in 1D, and seeing exactly how they fail here motivates the renormalization group. And through $T=e^{-\tau\hat H}$ the method seeds quantum Monte Carlo, the Suzuki–Trotter mapping, and the density-matrix renormalization group (DMRG) and modern tensor-network methods, both of which build on the transfer-matrix viewpoint.

The physical picture

Think of building the chain one spin at a time. Each new spin brings a Boltzmann factor that depends only on it and its left neighbour, so the only information you must carry forward is the value of the current spin — two possibilities. The transfer matrix $T$ is the $2\times2$ table of weights for the transition (old spin) $\to$ (new spin). Multiply $N$ such tables and take a trace to close the ring, and you have the exact partition function. All the physics is now the spectrum of one $2\times2$ matrix.

Figure

Building the chain: each bond contributes one transfer matrix, and the ring closes into a trace

The classical 1D chain is the imaginary-time history of a single quantum spin. The position along the chain plays the role of imaginary time $\tau$ , and $T=e^{-\tau_0\hat H}$ for an effective single-site quantum Hamiltonian. Ground-state dominance of $e^{-\tau\hat H}$ at large $\tau$ is exactly why only $\lambda_+$ matters at large $N$ ; the energy gap to the first excited state is $\ln(\lambda_+/\lambda_-)$ , i.e. the inverse correlation length. This dictionary — a $d$ -dimensional classical problem as a $(d-1)$ -dimensional quantum one — is one of the most productive ideas in the subject, and the 1D Ising chain is its simplest worked example. Why does the chain refuse to order? Start from the perfect ground state, all spins up. Flip a block of spins and you create two domain walls, each costing energy $2J$ no matter how far apart they sit. But the walls can be placed anywhere along $N$ sites, giving entropy $\sim2k_B\ln N$ . The free-energy change $\Delta F\sim4J-2k_BT\ln N$ turns negative for any $T>0$ once $N$ is large enough: disorder is thermodynamically inevitable. Energy is local, entropy is extensive, and in one dimension entropy always wins. The transfer-matrix spectrum says the same thing without pictures — the two eigenvalues stay split for all $T>0$ and only merge at $T=0$ .

Figure

Eigenvalue ratio and correlation length: gap stays open for all T＞0, closing only at T=0

The core equations

The one-dimensional Ising Hamiltonian on a periodic ring couples nearest neighbours and each spin to a field.

H = -J\sum_{i=1}^{N} s_i s_{i+1} - h\sum_{i=1}^{N} s_i,\qquad s_i=\pm1,\ s_{N+1}\equiv s_1 \qquad (1)

where: $J>0$ is the ferromagnetic nearest-neighbour exchange; $h$ the external field (energy units, absorbing $\mu_B$ and $g$ ); $s_i=\pm1$ the spin at site $i$ ; $N$ the number of sites; the periodic boundary closes the ring. Physically: every bond is identical, which is exactly what makes the trace trick work. Because the Hamiltonian is a sum of nearest-neighbour bond contributions, the Boltzmann weight factorizes into a product over bonds. Splitting the field contribution equally between the two neighbouring bonds of each spin makes the transfer matrix symmetric.

Z_N=\sum_{\{s\}}\prod_{i=1}^{N} T_{s_i,s_{i+1}},\qquad T_{s,s'}=e^{\,\beta J s s' + \tfrac{\beta h}{2}(s+s')} \qquad (2)

where: $\beta=1/k_BT$ ; $T_{s,s'}$ is the Boltzmann weight for an adjacent pair, with the field shared so $T$ is symmetric. Physically: the symmetric split does not change the physics but makes $T$ symmetric, hence real-diagonalizable with orthonormal eigenvectors. Written out in the ordered basis $(+1,-1)$ , the transfer matrix is a $2\times2$ table of exponentials.

T=\begin{pmatrix} e^{\beta J+\beta h} & e^{-\beta J} \\ e^{-\beta J} & e^{\beta J-\beta h}\end{pmatrix} \qquad (3)

where: the diagonal entries reward aligned pairs in the field, the off-diagonal entries $e^{-\beta J}$ penalize anti-aligned pairs. Physically: all four entries are strictly positive at finite $\beta$ — the property that forbids a 1D transition. Diagonalizing $T$ gives two eigenvalues; the larger controls the free energy, the smaller the correlation length.

\lambda_\pm = e^{\beta J}\cosh\beta h \pm \sqrt{e^{2\beta J}\sinh^2\beta h + e^{-2\beta J}} \qquad (4)

where: $\lambda_+$ is the largest eigenvalue (dominates the thermodynamic limit); $\lambda_-$ the subdominant one (sets $\xi$ ). Physically: the discriminant $e^{2\beta J}\sinh^2\beta h+e^{-2\beta J}\ge e^{-2\beta J}>0$ never vanishes at finite $T$ , so $\lambda_+$ is a smooth positive function everywhere. The partition function is the trace of $T^N$ , which in the thermodynamic limit is dominated by the largest eigenvalue alone.

Z_N=\mathrm{Tr}\,T^N=\lambda_+^N+\lambda_-^N \;\xrightarrow[N\to\infty]{}\; \lambda_+^N\!\left[1+(\lambda_-/\lambda_+)^N\right] \qquad (5)

where: the correction $(\lambda_-/\lambda_+)^N\to0$ because $0<\lambda_-/\lambda_+<1$ . Physically: the free energy per spin is set by $\lambda_+$ only, and since $\lambda_+$ is analytic there is no singularity — no transition. The free energy per spin follows immediately, and its smoothness is the whole story.

f=-k_BT\lim_{N\to\infty}\frac{\ln Z_N}{N}=-k_BT\ln\lambda_+ \qquad (6)

where: $f$ is the Helmholtz free energy per site. Physically: $\ln$ of a positive analytic function is analytic, so every derivative — magnetization, susceptibility, specific heat — is finite and smooth at all $T>0$ . Differentiating $f$ with respect to the field gives the magnetization, which vanishes at zero field for every finite temperature.

m=-\frac{\partial f}{\partial h}=\frac{\sinh\beta h}{\sqrt{\sinh^2\beta h+e^{-4\beta J}}} \qquad (7)

where: $m=\langle s_i\rangle$ is the magnetization per spin. Physically: at $h=0$ the numerator vanishes while the denominator stays finite ( $e^{-4\beta J}>0$ ), so $m=0$ for all $T>0$ ; only as $T\to0$ with $h>0$ does $m\to\mathrm{sgn}(h)$ . At zero field the two-point correlation function decays exponentially with separation, and the decay constant defines the correlation length. Summing the connected correlations yields the magnetic susceptibility.

\langle s_0 s_r\rangle = (\tanh\beta J)^{\,r},\quad \xi=-\frac{1}{\ln\tanh\beta J},\quad \chi=\beta\,\frac{1+\tanh\beta J}{1-\tanh\beta J}=\beta\,e^{2\beta J} \qquad (8)

where: $r$ is the separation in bonds; $\xi$ the correlation length ( $\langle s_0s_r\rangle\sim e^{-r/\xi}$ ); $\chi$ the zero-field susceptibility per spin. Physically: $\xi$ and $\chi$ both diverge exponentially as $T\to0$ , the mark of a critical point sitting exactly at zero temperature.

Derivation ASpin–spin correlation from the transfer-matrix eigenvectors

The goal is to compute the zero-field two-point function $\langle s_0s_r\rangle$ exactly and read off the correlation length from the ratio of the transfer-matrix eigenvalues. We insert the spin operator $\sigma^z=\mathrm{diag}(1,-1)$ at two positions inside the trace that defines $Z_N$ , diagonalize $T$ at $h=0$ , and take the thermodynamic limit so that only the largest eigenvalue survives in the "background" while the inserted operators pick up the subdominant eigenvalue between them. The result is a clean geometric series whose ratio is $\lambda_-/\lambda_+=\tanh\beta J$ , giving $\langle s_0s_r\rangle=(\tanh\beta J)^r$ and $\xi=-1/\ln\tanh\beta J$ , a length that diverges only as $T\to0$ .

Step 1. Operator insertion: $\langle s_0s_r\rangle=Z_N^{-1}\sum_{\{s\}}s_0s_r\prod_iT_{s_is_{i+1}}=\mathrm{Tr}(\sigma^zT^r\sigma^zT^{N-r})/\mathrm{Tr}\,T^N$ . Why: each spin factor $s_i$ becomes the diagonal matrix $\sigma^z$ acting at that site between transfer matrices.

Step 2. Diagonalize at $h=0$ : $T=e^{\beta J}I+e^{-\beta J}\sigma^x$ , with eigenvectors $|\pm\rangle=\tfrac1{\sqrt2}(1,\pm1)^{\!\top}$ and eigenvalues $\lambda_+=2\cosh\beta J$ , $\lambda_-=2\sinh\beta J$ . Why: adding a multiple of $\sigma^x$ to the identity is diagonal in the $|\pm\rangle$ basis.

Step 3. Take the traces with $N\to\infty$ : only the cross term survives, $\langle s_0s_r\rangle=\langle+|\sigma^z|-\rangle\langle-|\sigma^z|+\rangle\,(\lambda_-/\lambda_+)^r$ , and $\langle+|\sigma^z|-\rangle=1$ . Why: the background $\lambda_+^{N}$ cancels, leaving the subdominant eigenvalue raised to the separation $r$ .

Step 4. Identify the ratio: $\lambda_-/\lambda_+=\tanh\beta J$ , so $\langle s_0s_r\rangle=(\tanh\beta J)^r=e^{-r/\xi}$ with $\xi^{-1}=\ln(\lambda_+/\lambda_-)=-\ln\tanh\beta J>0$ for all finite $T$ . Why: the log of the eigenvalue ratio is the energy gap of the transfer operator, which is the inverse correlation length. *The zero-field eigenvalues $\lambda_+(h=0)=2\cosh\beta J$ and $\lambda_-(h=0)=2\sinh\beta J$ are consistent with brute-force enumeration on small rings, and the limits $\xi\to0$ as $T\to\infty$ and $\xi\to\infty$ as $T\to0$ follow directly from $\xi=-1/\ln\tanh\beta J$ .*

Go deeper · Derivation B · 1 minNo transition: analyticity of the free energy from Perron–Frobenius

Derivation BNo transition: analyticity of the free energy from Perron–Frobenius

The goal is to prove that the free energy $f=-k_BT\ln\lambda_+$ is analytic in $T$ and $h$ for every $T>0$ , so that no thermodynamic singularity — and hence no phase transition — can occur at finite temperature in the 1D chain. We use that the transfer matrix has strictly positive entries at finite $\beta$ , invoke the Perron–Frobenius theorem to conclude its largest eigenvalue is real, positive, nondegenerate and analytic in the entries, show the discriminant under the square root never vanishes, and finally note that the two eigenvalues become degenerate only as $T\to0$ — the transfer-matrix signature that the ordering "transition" sits exactly at zero temperature.

Step 1. Positivity. For every finite $\beta$ , all four entries of the transfer matrix $T$ are strictly positive. By Perron–Frobenius the largest eigenvalue $\lambda_+$ is real, positive, simple (nondegenerate), and an analytic function of the matrix entries. Why: a strictly positive matrix has a unique dominant eigenvalue that varies smoothly with its entries.

Step 2. No vanishing discriminant: $\Delta=e^{2\beta J}\sinh^2\beta h+e^{-2\beta J}\ge e^{-2\beta J}>0$ , even at $h=0$ where $\Delta=e^{-2\beta J}$ . Why: the square root stays real and strictly positive, so $\lambda_+$ has no branch-point singularity at any finite $T$ .

Step 3. Smooth free energy: $f(T,h)=-k_BT\ln\lambda_+(T,h)\in C^\infty$ , since the logarithm of a positive analytic function is analytic; every derivative (magnetization, susceptibility, specific heat) is finite. Why: a phase transition requires a non-analyticity in $f$ , and there is none.

Step 4. Degeneracy only at $T=0$ : $\lambda_-/\lambda_+=\tanh\beta J\to1$ as $T\to0$ at $h=0$ , so the two eigenvalues merge only there. Why: eigenvalue degeneracy of the transfer matrix is the fingerprint of symmetry breaking, and here it is pushed to zero temperature — the $T=0$ critical point. *One can confirm symbolically that $\Delta\ge e^{-2\beta J}>0$ for all finite $T$ , so $\lambda_+>\lambda_-$ at every $T>0$ , and that the ratio $\lambda_-/\lambda_+\to1$ as $T\to0$ , consistent with the eigenvalue degeneracy occurring only at zero temperature.*

Limiting cases

Behaviour of the $h=0$ chain in characteristic regimes (write $K=\beta J$ ):

Case	Result	Meaning
High temperature $K\to0$	$\xi\approx-1/\ln K\to0$ , $\langle s_0s_r\rangle\approx K^{\,r}$	Spins nearly independent; correlations die within one lattice spacing; $\chi\to$ Curie law $\sim1/T$ .
Low temperature $K\to\infty$	$\xi\approx\tfrac12 e^{2K}$ , $\chi\sim\beta e^{2\beta J}$	Correlation length and susceptibility diverge exponentially as $T\to0$ : a $T=0$ critical point.
Zero field, any $T>0$	$\langle s_i\rangle=0$ , $m=0$	No spontaneous magnetization; $f$ analytic; lower critical dimension of the Ising model is $d_\ell=1$ .
Zero temperature $T=0$	$\xi\to\infty$ , $m\to1$	The ground state is doubly degenerate and ordered. For the short-range one-dimensional Ising model, the only critical point occurs at $T=0$ .

Worked examples

Example 1 — nearest-neighbour correlation and bond energy at a definite temperature. For the zero-field chain at $K=\beta J=0.8$ , compute the nearest-neighbour correlation $\langle s_0s_1\rangle$ , the correlation length $\xi$ , and the internal energy per bond.

Step 1. Correlation: $\langle s_0s_1\rangle=\tanh K=\tanh(0.8)=\mathbf{0.6640}$ . Neighbouring spins are strongly but not perfectly correlated at this temperature.

Step 2. Correlation length: $\xi=-1/\ln\tanh K=-1/\ln(0.6640)=\mathbf{2.440}$ lattice spacings. Correlations survive a couple of bonds and then decay.

Step 3. Internal energy per bond: $u=-J\langle s_0s_1\rangle=-J\tanh K=\mathbf{-0.6640\,J}$ . As $T\to0$ ( $K\to\infty$ ) this approaches $-J$ (every bond satisfied); as $T\to\infty$ it approaches $0$ (bonds random).

Example 2 — the correlation length diverges only at zero temperature. Track $\xi(T)$ as $T$ falls and confirm it blows up exponentially rather than at any finite temperature.

Step 1. At $K=1$ : $\tanh1=0.7616$ , $\xi=-1/\ln(0.7616)=\mathbf{3.667}$ .

Step 2. At $K=2$ : $\tanh2=0.9640$ , $\xi=-1/\ln(0.9640)=\mathbf{27.3}$ . Doubling $K=\beta J$ (equivalently, halving the temperature) increases the correlation length by more than a factor of seven.

Step 3. The asymptotic low-temperature expression $\xi \simeq \frac12 e^{2K}$ gives $\xi=\tfrac12 e^4=27.3$ at $K=2$ , in excellent agreement with the exact result from Step 2. Since $\xi$ grows as $e^{2\beta J}$ , it is finite at every $T>0$ and diverges only as $T\to0$ — there is no finite-temperature transition, exactly as the analyticity of $\lambda_+$ requires.

Origins · 1 mintransfer matrix and how this was discovered

transfer matrix

method — Kramers & Wannier 1941 · The transfer matrix is the central technique for exactly solvable lattice models, introduced by Hendrik Kramers and Gregory Wannier in 1941 in their study of the two-dimensional Ising model. Their insight was that a statistical partition function with local (nearest-neighbour) interactions can be rewritten as the trace of a product of identical matrices — one per row or bond — so that the thermodynamics is encoded entirely in the matrix's largest eigenvalue, and the correlation length in the ratio of its two largest eigenvalues. For the 1D chain the matrix is a humble 2×2, and its analyticity proves at once that there is no finite-temperature order. For the 2D model the transfer matrix becomes 2^L-dimensional for a strip of width L, and it was by diagonalizing exactly this object that Lars Onsager achieved his 1944 solution. The method also crystallized the deep quantum–classical correspondence — the transfer matrix is e^{−τĤ} for an effective quantum Hamiltonian — that underlies quantum Monte Carlo and the modern theory of tensor networks.

History

1920

— Lenz proposes the model of coupled binary spins

1925

— Ising solves the 1D chain, finds no phase transition, and leaves physics

1936

— Peierls gives the domain-wall argument for order in the 2D model

1941

— Kramers and Wannier introduce the transfer-matrix method and locate the 2D T_c by duality

1944

— Onsager diagonalizes the 2D transfer matrix exactly, giving the first nontrivial exact solution

1950

— the quantum-classical (transfer-matrix as e^{−τĤ}) correspondence is fully articulated

1976

— Suzuki–Trotter mapping turns d-dimensional quantum models into (d+1)-dimensional classical ones

Going further

References. Pathria & Beale, Statistical Mechanics, 3rd ed., §13.2 (One-dimensional Ising / transfer matrix) · Kardar, Statistical Physics of Fields, Ch. 6 (Transfer matrices) · Reif, Fundamentals of Statistical and Thermal Physics · Sethna, Statistical Mechanics: Entropy, Order Parameters, and Complexity, Ch. 8 · Baxter, Exactly Solved Models in Statistical Mechanics, Ch. 2 · Landau & Lifshitz, Statistical Physics Part 1, §163. The eigenvalues follow from the characteristic equation $\lambda^2-2e^{\beta J}\cosh\beta h\,\lambda+2\sinh2\beta J=0$ ; the magnetization $m=\sinh\beta h/\sqrt{\sinh^2\beta h+e^{-4\beta J}}$ from $m=-\partial f/\partial h$ ; the correlation $\langle s_0s_r\rangle=(\tanh\beta J)^r$ and $\xi=-1/\ln\tanh\beta J$ from the eigenvector insertion; and the susceptibility $\chi=\beta e^{2\beta J}$ from the geometric sum. The transfer-matrix method itself is due to Kramers & Wannier (1941).

Builds on: The Ising Model: Definition and Hamiltonian · Transfer Matrix Formalism for Lattice Models · Order Parameter and Spontaneous Magnetization. Leads to: Peierls Argument for Long-Range Order · Onsager Solution of the Two-Dimensional Ising Model · Kramers-Wannier Duality · Domain Walls and Interfacial Excitations.

What exactly is the transfer matrix, and why is it

2\times2

here?

It is the $2\times2$ table of Boltzmann weights $T_{s,s'}$ for one bond, connecting a spin $s$ to its neighbour $s'$ . Because the 1D energy is a sum over independent bonds, the total Boltzmann weight is a product of these matrices, and summing over all spin configurations with periodic boundaries is exactly the trace of $T^N$ . It is $2\times2$ because a single Ising spin has two states, so the "state" you carry from one site to the next is one of two values. In higher dimensions or with more spin states the matrix grows, but the logic is identical.

How do I get the correlation length from the eigenvalues?

The inverse correlation length is the gap between the two largest eigenvalues on a log scale: $\xi^{-1}=\ln(\lambda_+/\lambda_-)$ . At zero field the ratio is $\lambda_-/\lambda_+=\tanh\beta J$ , so $\xi=-1/\ln\tanh\beta J$ . Physically, correlations propagate along the chain by "exciting" the subdominant eigenvector, and the further you go the more suppressed that excitation is by the eigenvalue ratio — hence exponential decay $e^{-r/\xi}$ . The correlation length diverges precisely when the eigenvalues become degenerate, which for the 1D chain happens only at $T=0$ .

Why is there a phase transition at

T=0

but not at any

T>0

At exactly $T=0$ the two eigenvalues become degenerate, $\lambda_-/\lambda_+=\tanh\beta J\to1$ , so the correlation length diverges and the ground state (all up or all down) is genuinely ordered and degenerate. This is a bona fide critical point — but a special one, sitting at the edge of the accessible temperature range. For any $T>0$ the eigenvalues are split, $f$ is analytic, and there is no singularity. So the 1D Ising model has its lower critical dimension exactly at $d=1$ : order barely fails to survive to finite temperature.

Does the 1D result mean the Ising model is useless?

Not at all — that was Ising's own mistaken conclusion in 1925. The 1D chain is exactly soluble and teaches the transfer-matrix method, the energy–entropy balance that fixes the lower critical dimension, and the quantum–classical correspondence. Moreover the physics changes completely in two dimensions, where Onsager's exact solution shows a genuine finite-temperature transition with non-classical exponents. The 1D result is best read as a statement about dimension: whether order survives thermal fluctuations depends on dimensionality, and one dimension is simply too "floppy" for a discrete symmetry to stay broken at $T>0$ .

What is the quantum–classical mapping and why does it matter?

The transfer matrix can be written as $T=e^{-\tau_0\hat H}$ for an effective single-site quantum Hamiltonian, so marching along the classical chain is exactly evolving a quantum spin in imaginary time. This means a $d$ -dimensional classical statistical problem is equivalent to a $(d-1)$ -dimensional quantum one (and, run the other way, a $d$ -dimensional quantum problem maps to a $(d+1)$ -dimensional classical one via Suzuki–Trotter). The 1D Ising chain is the simplest example: it corresponds to a single qubit in imaginary time. This dictionary underlies quantum Monte Carlo, path-integral methods, and tensor-network algorithms.

Why symmetrize the field between the two bonds?

Splitting each site's field energy $-hs_i$ equally between its two bonds makes the transfer matrix symmetric, $T_{s,s'}=T_{s',s}$ . A symmetric matrix has real eigenvalues and orthonormal eigenvectors, which makes the correlation-function calculation clean and guarantees the eigenvalues behave nicely. You could instead put the whole field term on one side of each bond; the eigenvalues and all physical quantities are unchanged, but $T$ would be non-symmetric and the algebra messier. The symmetric split is a convenience, not a physical choice.